Hi Matt,
Are you using an event-related or block design (i.e. are your events'
durations 0 or not?)?
Best regards,
Guillaume.
On 04/09/2019 20:30, Heard, Matt wrote:
> Hello SPM users,
>
> The short version of my question is: would a 3x1 within-subject ANOVA
> pick up on gross differences in beta magnitudes across factors?
>
> The long version of my question is: I would like to perform a 3x1
> repeated measures ANOVA wherein the factor of interest is the
> acquisition protocol used to acquire data. In essence, I have fourteen
> subjects who performed a task while being scanned with three different
> scanning parameters:
>
> 1. Paradigm 1: TR of 1 second, duration ~5 minutes per run.
> 2. Paradigm 2: TR of 2 seconds, duration ~5 minutes per run.
> 3. Paradigm 3: TR of 1 second, duration ~5 minutes per run.
>
> The experimental design was task-related and each run involved the same
> number of events. One of the most salient differences in the data is
> that paradigms 1 and 3 acquired twice as much data (better temporal
> resolution) than paradigm 2.
>
> Thus far I have completed first-level analysis of each subject during
> each scan and have created three GLMs per subject. I noticed that the
> magnitude of betas in paradigms 1 and 3 (with TR of 1 second) are about
> half as large as in paradigm 2 (TR of 2 seconds). I suspected that this
> was a quirk of statistics due to the mismatched TR/decreased amount of
> data collected, so I verified this by redoing the analysis of paradigm 1
> and down-sampled the data to match the size/temporal resolution of
> paradigm 2. The magnitude of beta estimates now matched between the
> down-sampled paradigm 1 and paradigm 2 models.
>
> I am unsure which SPM output file to submit to the ANOVA--con or spmT.
> With all other second-level analyses I have been informed that con files
> are appropriate, so I assume the same applies here. However, the
> difference in beta estimates (due to amount of data/temporal resolution)
> makes me pause. *Could the ANOVA results pick up on gross differences in
> beta magnitude across the scans, thus resulting in an obvious effect of
> scan that may not represent differences in signal/noise across the three
> scanning paradigms? If so, how could I perform an ANOVA that is agnostic
> to differences in beta magnitude?
> *
>
> Thanks,
>
> Matthew Heard
> Graduate Student
> *The Ohio State University*
> Arts and Sciences Neuroscience Graduate Program
> 004 Pressey Hall, 1070 Carmack Road, Columbus, OH 43201
> 214-458-7255 Mobile
> [log in to unmask] <mailto:[log in to unmask]>
> <https://eur01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fosu.edu&data=02%7C01%7C%7Cf5c7fd86fbab42fd13a408d7316e4f86%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637032222160757814&sdata=fLHOJFH4nVrdf2Vp3DVMI450P2CLTNGMsjHGazLFLoc%3D&reserved=0>https://osu-slam.github.io/
> <https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fosu-slam.github.io%2F&data=02%7C01%7C%7Cf5c7fd86fbab42fd13a408d7316e4f86%7C1faf88fea9984c5b93c9210a11d9a5c2%7C0%7C0%7C637032222160757814&sdata=yfxNG5bs8UAA24gKPSCWd3hBJNeJESHUOJSGTbOAszc%3D&reserved=0>
>
> /Buckeyes consider the environment before printing./
>
>
--
Guillaume Flandin, PhD
Wellcome Centre for Human Neuroimaging
UCL Queen Square Institute of Neurology
London WC1N 3BG
|